Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1/*
2 * Linux Socket Filter Data Structures
3 */
4#ifndef __LINUX_FILTER_H__
5#define __LINUX_FILTER_H__
6
7#include <stdarg.h>
8
9#include <linux/atomic.h>
10#include <linux/compat.h>
11#include <linux/skbuff.h>
12#include <linux/linkage.h>
13#include <linux/printk.h>
14#include <linux/workqueue.h>
15#include <linux/sched.h>
16
17#include <asm/cacheflush.h>
18
19#include <uapi/linux/filter.h>
20#include <uapi/linux/bpf.h>
21
22struct sk_buff;
23struct sock;
24struct seccomp_data;
25struct bpf_prog_aux;
26
27/* ArgX, context and stack frame pointer register positions. Note,
28 * Arg1, Arg2, Arg3, etc are used as argument mappings of function
29 * calls in BPF_CALL instruction.
30 */
31#define BPF_REG_ARG1 BPF_REG_1
32#define BPF_REG_ARG2 BPF_REG_2
33#define BPF_REG_ARG3 BPF_REG_3
34#define BPF_REG_ARG4 BPF_REG_4
35#define BPF_REG_ARG5 BPF_REG_5
36#define BPF_REG_CTX BPF_REG_6
37#define BPF_REG_FP BPF_REG_10
38
39/* Additional register mappings for converted user programs. */
40#define BPF_REG_A BPF_REG_0
41#define BPF_REG_X BPF_REG_7
42#define BPF_REG_TMP BPF_REG_8
43
44/* BPF program can access up to 512 bytes of stack space. */
45#define MAX_BPF_STACK 512
46
47/* Helper macros for filter block array initializers. */
48
49/* ALU ops on registers, bpf_add|sub|...: dst_reg += src_reg */
50
51#define BPF_ALU64_REG(OP, DST, SRC) \
52 ((struct bpf_insn) { \
53 .code = BPF_ALU64 | BPF_OP(OP) | BPF_X, \
54 .dst_reg = DST, \
55 .src_reg = SRC, \
56 .off = 0, \
57 .imm = 0 })
58
59#define BPF_ALU32_REG(OP, DST, SRC) \
60 ((struct bpf_insn) { \
61 .code = BPF_ALU | BPF_OP(OP) | BPF_X, \
62 .dst_reg = DST, \
63 .src_reg = SRC, \
64 .off = 0, \
65 .imm = 0 })
66
67/* ALU ops on immediates, bpf_add|sub|...: dst_reg += imm32 */
68
69#define BPF_ALU64_IMM(OP, DST, IMM) \
70 ((struct bpf_insn) { \
71 .code = BPF_ALU64 | BPF_OP(OP) | BPF_K, \
72 .dst_reg = DST, \
73 .src_reg = 0, \
74 .off = 0, \
75 .imm = IMM })
76
77#define BPF_ALU32_IMM(OP, DST, IMM) \
78 ((struct bpf_insn) { \
79 .code = BPF_ALU | BPF_OP(OP) | BPF_K, \
80 .dst_reg = DST, \
81 .src_reg = 0, \
82 .off = 0, \
83 .imm = IMM })
84
85/* Endianess conversion, cpu_to_{l,b}e(), {l,b}e_to_cpu() */
86
87#define BPF_ENDIAN(TYPE, DST, LEN) \
88 ((struct bpf_insn) { \
89 .code = BPF_ALU | BPF_END | BPF_SRC(TYPE), \
90 .dst_reg = DST, \
91 .src_reg = 0, \
92 .off = 0, \
93 .imm = LEN })
94
95/* Short form of mov, dst_reg = src_reg */
96
97#define BPF_MOV64_REG(DST, SRC) \
98 ((struct bpf_insn) { \
99 .code = BPF_ALU64 | BPF_MOV | BPF_X, \
100 .dst_reg = DST, \
101 .src_reg = SRC, \
102 .off = 0, \
103 .imm = 0 })
104
105#define BPF_MOV32_REG(DST, SRC) \
106 ((struct bpf_insn) { \
107 .code = BPF_ALU | BPF_MOV | BPF_X, \
108 .dst_reg = DST, \
109 .src_reg = SRC, \
110 .off = 0, \
111 .imm = 0 })
112
113/* Short form of mov, dst_reg = imm32 */
114
115#define BPF_MOV64_IMM(DST, IMM) \
116 ((struct bpf_insn) { \
117 .code = BPF_ALU64 | BPF_MOV | BPF_K, \
118 .dst_reg = DST, \
119 .src_reg = 0, \
120 .off = 0, \
121 .imm = IMM })
122
123#define BPF_MOV32_IMM(DST, IMM) \
124 ((struct bpf_insn) { \
125 .code = BPF_ALU | BPF_MOV | BPF_K, \
126 .dst_reg = DST, \
127 .src_reg = 0, \
128 .off = 0, \
129 .imm = IMM })
130
131/* BPF_LD_IMM64 macro encodes single 'load 64-bit immediate' insn */
132#define BPF_LD_IMM64(DST, IMM) \
133 BPF_LD_IMM64_RAW(DST, 0, IMM)
134
135#define BPF_LD_IMM64_RAW(DST, SRC, IMM) \
136 ((struct bpf_insn) { \
137 .code = BPF_LD | BPF_DW | BPF_IMM, \
138 .dst_reg = DST, \
139 .src_reg = SRC, \
140 .off = 0, \
141 .imm = (__u32) (IMM) }), \
142 ((struct bpf_insn) { \
143 .code = 0, /* zero is reserved opcode */ \
144 .dst_reg = 0, \
145 .src_reg = 0, \
146 .off = 0, \
147 .imm = ((__u64) (IMM)) >> 32 })
148
149/* pseudo BPF_LD_IMM64 insn used to refer to process-local map_fd */
150#define BPF_LD_MAP_FD(DST, MAP_FD) \
151 BPF_LD_IMM64_RAW(DST, BPF_PSEUDO_MAP_FD, MAP_FD)
152
153/* Short form of mov based on type, BPF_X: dst_reg = src_reg, BPF_K: dst_reg = imm32 */
154
155#define BPF_MOV64_RAW(TYPE, DST, SRC, IMM) \
156 ((struct bpf_insn) { \
157 .code = BPF_ALU64 | BPF_MOV | BPF_SRC(TYPE), \
158 .dst_reg = DST, \
159 .src_reg = SRC, \
160 .off = 0, \
161 .imm = IMM })
162
163#define BPF_MOV32_RAW(TYPE, DST, SRC, IMM) \
164 ((struct bpf_insn) { \
165 .code = BPF_ALU | BPF_MOV | BPF_SRC(TYPE), \
166 .dst_reg = DST, \
167 .src_reg = SRC, \
168 .off = 0, \
169 .imm = IMM })
170
171/* Direct packet access, R0 = *(uint *) (skb->data + imm32) */
172
173#define BPF_LD_ABS(SIZE, IMM) \
174 ((struct bpf_insn) { \
175 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_ABS, \
176 .dst_reg = 0, \
177 .src_reg = 0, \
178 .off = 0, \
179 .imm = IMM })
180
181/* Indirect packet access, R0 = *(uint *) (skb->data + src_reg + imm32) */
182
183#define BPF_LD_IND(SIZE, SRC, IMM) \
184 ((struct bpf_insn) { \
185 .code = BPF_LD | BPF_SIZE(SIZE) | BPF_IND, \
186 .dst_reg = 0, \
187 .src_reg = SRC, \
188 .off = 0, \
189 .imm = IMM })
190
191/* Memory load, dst_reg = *(uint *) (src_reg + off16) */
192
193#define BPF_LDX_MEM(SIZE, DST, SRC, OFF) \
194 ((struct bpf_insn) { \
195 .code = BPF_LDX | BPF_SIZE(SIZE) | BPF_MEM, \
196 .dst_reg = DST, \
197 .src_reg = SRC, \
198 .off = OFF, \
199 .imm = 0 })
200
201/* Memory store, *(uint *) (dst_reg + off16) = src_reg */
202
203#define BPF_STX_MEM(SIZE, DST, SRC, OFF) \
204 ((struct bpf_insn) { \
205 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_MEM, \
206 .dst_reg = DST, \
207 .src_reg = SRC, \
208 .off = OFF, \
209 .imm = 0 })
210
211/* Atomic memory add, *(uint *)(dst_reg + off16) += src_reg */
212
213#define BPF_STX_XADD(SIZE, DST, SRC, OFF) \
214 ((struct bpf_insn) { \
215 .code = BPF_STX | BPF_SIZE(SIZE) | BPF_XADD, \
216 .dst_reg = DST, \
217 .src_reg = SRC, \
218 .off = OFF, \
219 .imm = 0 })
220
221/* Memory store, *(uint *) (dst_reg + off16) = imm32 */
222
223#define BPF_ST_MEM(SIZE, DST, OFF, IMM) \
224 ((struct bpf_insn) { \
225 .code = BPF_ST | BPF_SIZE(SIZE) | BPF_MEM, \
226 .dst_reg = DST, \
227 .src_reg = 0, \
228 .off = OFF, \
229 .imm = IMM })
230
231/* Conditional jumps against registers, if (dst_reg 'op' src_reg) goto pc + off16 */
232
233#define BPF_JMP_REG(OP, DST, SRC, OFF) \
234 ((struct bpf_insn) { \
235 .code = BPF_JMP | BPF_OP(OP) | BPF_X, \
236 .dst_reg = DST, \
237 .src_reg = SRC, \
238 .off = OFF, \
239 .imm = 0 })
240
241/* Conditional jumps against immediates, if (dst_reg 'op' imm32) goto pc + off16 */
242
243#define BPF_JMP_IMM(OP, DST, IMM, OFF) \
244 ((struct bpf_insn) { \
245 .code = BPF_JMP | BPF_OP(OP) | BPF_K, \
246 .dst_reg = DST, \
247 .src_reg = 0, \
248 .off = OFF, \
249 .imm = IMM })
250
251/* Function call */
252
253#define BPF_EMIT_CALL(FUNC) \
254 ((struct bpf_insn) { \
255 .code = BPF_JMP | BPF_CALL, \
256 .dst_reg = 0, \
257 .src_reg = 0, \
258 .off = 0, \
259 .imm = ((FUNC) - __bpf_call_base) })
260
261/* Raw code statement block */
262
263#define BPF_RAW_INSN(CODE, DST, SRC, OFF, IMM) \
264 ((struct bpf_insn) { \
265 .code = CODE, \
266 .dst_reg = DST, \
267 .src_reg = SRC, \
268 .off = OFF, \
269 .imm = IMM })
270
271/* Program exit */
272
273#define BPF_EXIT_INSN() \
274 ((struct bpf_insn) { \
275 .code = BPF_JMP | BPF_EXIT, \
276 .dst_reg = 0, \
277 .src_reg = 0, \
278 .off = 0, \
279 .imm = 0 })
280
281/* Internal classic blocks for direct assignment */
282
283#define __BPF_STMT(CODE, K) \
284 ((struct sock_filter) BPF_STMT(CODE, K))
285
286#define __BPF_JUMP(CODE, K, JT, JF) \
287 ((struct sock_filter) BPF_JUMP(CODE, K, JT, JF))
288
289#define bytes_to_bpf_size(bytes) \
290({ \
291 int bpf_size = -EINVAL; \
292 \
293 if (bytes == sizeof(u8)) \
294 bpf_size = BPF_B; \
295 else if (bytes == sizeof(u16)) \
296 bpf_size = BPF_H; \
297 else if (bytes == sizeof(u32)) \
298 bpf_size = BPF_W; \
299 else if (bytes == sizeof(u64)) \
300 bpf_size = BPF_DW; \
301 \
302 bpf_size; \
303})
304
305/* Macro to invoke filter function. */
306#define SK_RUN_FILTER(filter, ctx) \
307 (*filter->prog->bpf_func)(ctx, filter->prog->insnsi)
308
309#ifdef CONFIG_COMPAT
310/* A struct sock_filter is architecture independent. */
311struct compat_sock_fprog {
312 u16 len;
313 compat_uptr_t filter; /* struct sock_filter * */
314};
315#endif
316
317struct sock_fprog_kern {
318 u16 len;
319 struct sock_filter *filter;
320};
321
322struct bpf_binary_header {
323 unsigned int pages;
324 u8 image[];
325};
326
327struct bpf_prog {
328 u16 pages; /* Number of allocated pages */
329 bool jited; /* Is our filter JIT'ed? */
330 bool gpl_compatible; /* Is our filter GPL compatible? */
331 u32 len; /* Number of filter blocks */
332 enum bpf_prog_type type; /* Type of BPF program */
333 struct bpf_prog_aux *aux; /* Auxiliary fields */
334 struct sock_fprog_kern *orig_prog; /* Original BPF program */
335 unsigned int (*bpf_func)(const struct sk_buff *skb,
336 const struct bpf_insn *filter);
337 /* Instructions for interpreter */
338 union {
339 struct sock_filter insns[0];
340 struct bpf_insn insnsi[0];
341 };
342};
343
344struct sk_filter {
345 atomic_t refcnt;
346 struct rcu_head rcu;
347 struct bpf_prog *prog;
348};
349
350#define BPF_PROG_RUN(filter, ctx) (*filter->bpf_func)(ctx, filter->insnsi)
351
352static inline unsigned int bpf_prog_size(unsigned int proglen)
353{
354 return max(sizeof(struct bpf_prog),
355 offsetof(struct bpf_prog, insns[proglen]));
356}
357
358static inline bool bpf_prog_was_classic(const struct bpf_prog *prog)
359{
360 /* When classic BPF programs have been loaded and the arch
361 * does not have a classic BPF JIT (anymore), they have been
362 * converted via bpf_migrate_filter() to eBPF and thus always
363 * have an unspec program type.
364 */
365 return prog->type == BPF_PROG_TYPE_UNSPEC;
366}
367
368#define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
369
370#ifdef CONFIG_DEBUG_SET_MODULE_RONX
371static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
372{
373 set_memory_ro((unsigned long)fp, fp->pages);
374}
375
376static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
377{
378 set_memory_rw((unsigned long)fp, fp->pages);
379}
380#else
381static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
382{
383}
384
385static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
386{
387}
388#endif /* CONFIG_DEBUG_SET_MODULE_RONX */
389
390int sk_filter(struct sock *sk, struct sk_buff *skb);
391
392int bpf_prog_select_runtime(struct bpf_prog *fp);
393void bpf_prog_free(struct bpf_prog *fp);
394
395struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags);
396struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
397 gfp_t gfp_extra_flags);
398void __bpf_prog_free(struct bpf_prog *fp);
399
400static inline void bpf_prog_unlock_free(struct bpf_prog *fp)
401{
402 bpf_prog_unlock_ro(fp);
403 __bpf_prog_free(fp);
404}
405
406typedef int (*bpf_aux_classic_check_t)(struct sock_filter *filter,
407 unsigned int flen);
408
409int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog);
410int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
411 bpf_aux_classic_check_t trans);
412void bpf_prog_destroy(struct bpf_prog *fp);
413
414int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk);
415int sk_attach_bpf(u32 ufd, struct sock *sk);
416int sk_detach_filter(struct sock *sk);
417int sk_get_filter(struct sock *sk, struct sock_filter __user *filter,
418 unsigned int len);
419
420bool sk_filter_charge(struct sock *sk, struct sk_filter *fp);
421void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp);
422
423u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
424void bpf_int_jit_compile(struct bpf_prog *fp);
425bool bpf_helper_changes_skb_data(void *func);
426
427#ifdef CONFIG_BPF_JIT
428typedef void (*bpf_jit_fill_hole_t)(void *area, unsigned int size);
429
430struct bpf_binary_header *
431bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
432 unsigned int alignment,
433 bpf_jit_fill_hole_t bpf_fill_ill_insns);
434void bpf_jit_binary_free(struct bpf_binary_header *hdr);
435
436void bpf_jit_compile(struct bpf_prog *fp);
437void bpf_jit_free(struct bpf_prog *fp);
438
439static inline void bpf_jit_dump(unsigned int flen, unsigned int proglen,
440 u32 pass, void *image)
441{
442 pr_err("flen=%u proglen=%u pass=%u image=%pK from=%s pid=%d\n", flen,
443 proglen, pass, image, current->comm, task_pid_nr(current));
444
445 if (image)
446 print_hex_dump(KERN_ERR, "JIT code: ", DUMP_PREFIX_OFFSET,
447 16, 1, image, proglen, false);
448}
449#else
450static inline void bpf_jit_compile(struct bpf_prog *fp)
451{
452}
453
454static inline void bpf_jit_free(struct bpf_prog *fp)
455{
456 bpf_prog_unlock_free(fp);
457}
458#endif /* CONFIG_BPF_JIT */
459
460#define BPF_ANC BIT(15)
461
462static inline u16 bpf_anc_helper(const struct sock_filter *ftest)
463{
464 BUG_ON(ftest->code & BPF_ANC);
465
466 switch (ftest->code) {
467 case BPF_LD | BPF_W | BPF_ABS:
468 case BPF_LD | BPF_H | BPF_ABS:
469 case BPF_LD | BPF_B | BPF_ABS:
470#define BPF_ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE: \
471 return BPF_ANC | SKF_AD_##CODE
472 switch (ftest->k) {
473 BPF_ANCILLARY(PROTOCOL);
474 BPF_ANCILLARY(PKTTYPE);
475 BPF_ANCILLARY(IFINDEX);
476 BPF_ANCILLARY(NLATTR);
477 BPF_ANCILLARY(NLATTR_NEST);
478 BPF_ANCILLARY(MARK);
479 BPF_ANCILLARY(QUEUE);
480 BPF_ANCILLARY(HATYPE);
481 BPF_ANCILLARY(RXHASH);
482 BPF_ANCILLARY(CPU);
483 BPF_ANCILLARY(ALU_XOR_X);
484 BPF_ANCILLARY(VLAN_TAG);
485 BPF_ANCILLARY(VLAN_TAG_PRESENT);
486 BPF_ANCILLARY(PAY_OFFSET);
487 BPF_ANCILLARY(RANDOM);
488 BPF_ANCILLARY(VLAN_TPID);
489 }
490 /* Fallthrough. */
491 default:
492 return ftest->code;
493 }
494}
495
496void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb,
497 int k, unsigned int size);
498
499static inline void *bpf_load_pointer(const struct sk_buff *skb, int k,
500 unsigned int size, void *buffer)
501{
502 if (k >= 0)
503 return skb_header_pointer(skb, k, size, buffer);
504
505 return bpf_internal_load_pointer_neg_helper(skb, k, size);
506}
507
508static inline int bpf_tell_extensions(void)
509{
510 return SKF_AD_MAX;
511}
512
513#endif /* __LINUX_FILTER_H__ */